README.md

metadata

license: apache-2.0
base_model: MongoDB/mdbr-leaf-ir
tags:
  - coreml
  - sentence-similarity
  - feature-extraction

mdbr-leaf-ir, Core ML (fp16)

This is MongoDB's mdbr-leaf-ir (revision 4262131b, Apache License 2.0), converted to Core ML by Innerloop for Breadcrumb's search and recall by meaning. All credit for the model goes to MongoDB.

• leaf.mlmodelc: the compiled Core ML program, fp16 storage, macOS 14+. Input input_ids (int32, shape [1, L], L one of 16, 32, 64, 128, 256, 512; pad with 0). Output embedding: the finished 768-float, unit-length vector. The model's own mean pooling, its 384-to-768 dense layer and its normalisation are inside the program.
• vocab.txt: the model's own WordPiece vocabulary (BERT uncased), unchanged.
• Queries take the prefix Represent this sentence for searching relevant passages: ; documents take none.
• LICENSE (Apache 2.0) and NOTICE carry the attribution.